Correct in what way? Not correct if you want the best performing model. And if it doesn't outperform the current best model nobody will care about you. For a company like Mistral that could be the end for them.

I'm not saying it uses books3, it might not. I'm just saying why it might make sense to risk it.

Nah. Copyright law isn't compatible with small startups and independent developers being able to compete with megacorps in AI. It's unironically better to just ignore it.

You seem to be unfamiliar with the mantra of silicon valley.